If I'm manipulating the statistics in my head properly, even with a skewed sample, Alexa stats should converge on the truth as you go higher in the ranking, because the distribution of web site accesses is very skewed towards the top websites, which should have the effect of limiting the error as you get closer to #1. If Alexa says a site is well within the top 1000 (as opposed to just on the border, and Alexa shows them at 771), I'd accept that as strong evidence.